GPUs & Macs › AMD Instinct MI300X
AMD Instinct MI300X: local LLM speed
Llama 2 7B (Q4_0) generates 233 tokens/s on the AMD Instinct MI300X (ROCm), and processes prompts at 11,476 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 233 | 11,476 | Q4_0 | ROCm FA off | tg128 | llama.cpp #15021 details |
| Llama 2 7B | 219 | 11,946 | Q4_0 | ROCm FA on | tg128 | llama.cpp #15021 details |
What fits on the AMD Instinct MI300X
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 233 tok/s (ROCm, source). Fit assumes the whole model in GPU memory with ~8K context.
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| Qwen3.8-27B | Q8_0 | 29.0 GB | 30.7 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 70.3 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 93.4 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 199 estimate |
| Qwen3.6-35B-A3B | Q8_0 | 36.9 GB | 141 estimate |
| Qwen3.8-Flash-Next | Q8_0 | 188.2 GB | 4.7 estimate |
| gemma-4-26B-A4B-it | Q8_0 | 26.9 GB | 108 estimate |
| GLM-5.3-Flash | UD-IQ4_XS | 156.8 GB | 5.7 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 93.3 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 109 estimate |
| Muse-Glimmer-30B | Q8_0 | 29.6 GB | 30.1 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 208 estimate |
| gemma-4-31B-it | Q8_0 | 32.6 GB | 27.3 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 203 estimate |
Similar hardware
AMD Instinct MI210 · AMD Instinct MI50 · AMD Instinct MI60 · AMD Instinct MI100 · AMD Instinct MI25